Semantic Communication: Methods & Insights

• Semantic Communication (SemCom) is an innovative approach that transmits extracted meaning instead of raw bits, improving spectral efficiency and context-awareness.
• Hybrid transceiver designs enable simultaneous SemCom–BitCom operations by allocating power and employing dual-path encoding for diverse traffic profiles.
• Advanced techniques like reinforcement learning and multi-modal fusion optimize resource allocation and system performance in constrained and non-terrestrial environments.

Semantic Communication (SemCom) is an advanced communication paradigm targeting the transmission of extracted meaning—rather than raw data—across physical channels. By focusing on semantics, SemCom offers substantial gains in spectral efficiency, resource utilization, and context-awareness, especially in bandwidth and power-constrained environments. This approach, which distinctively departs from classic bit-centric models, is central to the evolution of next-generation (6G+) wireless networks, heterogeneous multiple-access, and intelligent distributed systems.

1. Foundational Concepts and Formalization

Semantic Communication redefines the communication objective. Traditional Bit-based Communication (BitCom) seeks to reconstruct a bit-string $b \in \{0,1\}^M$ with an arbitrarily low bit-error probability, optimizing under Shannon’s channel-capacity constraint: $R_B \le W\log_2(1+\gamma)$, where $W$ is bandwidth and $\gamma$ is SNR. In contrast, SemCom frames the problem as extracting and transmitting a minimal representation $s$ that carries the intended meaning $u$, where only task-relevant information is conveyed and reconstructed at the destination (Ahmed et al., 13 Jun 2025). The underlying principle is that the metrics and architectures should target semantic-level accuracy, not bit-level fidelity.

Semantic-level metrics include:

• Semantic Entropy: $H_s(S) = -\sum_{s} P(s) \log P(s)$, quantifying uncertainty over the set of semantic symbols $S$.
• Semantic Distortion: $$D_s = E[d_s(S,\hat S)] = \sum_{s,\hat s} P(s)P(\hat s|s)d_s(s,\hat s)$$, measuring semantic loss under a metric $d_s$.
• Semantic Rate: The minimum rate $$R_s(D_s) = \inf_{P(\hat s|s):E[d_s]\le D_s} I(S;\hat S)$$ to achieve $D_s$ distortion.

Practical implementations such as DeepSC further operationalize this by using semantic similarity $\epsilon(K,\gamma) \in [0,1]$ and compute the semantic rate as $R_s = \frac{WI}{KL}\epsilon(K,\gamma)$, with $I$ semantic information units per message, $K$ symbols per word, and $L$ the number of words (Ahmed et al., 13 Jun 2025).

2. SemCom–BitCom Coexistence and Hybrid Transceiver Design

The coexistence of SemCom and BitCom is a key architectural innovation in heterogeneous networks, allowing simultaneous support for users and applications with divergent communication needs. In such a system, transmitters implement a dual-path design:

• Bit Path: $$b \rightarrow f_B(b) \rightarrow x_B \in \mathbb{C}^n$$
• Semantic Path: $$u \rightarrow f_S(u) \rightarrow S \in \mathbb{R}^k \rightarrow c_S(S) \rightarrow x_S$$

The hybrid superposition signal transmitted over the channel is:

$x = \sqrt{p_B}x_B + \sqrt{p_S}x_S$

or in a power-fraction representation:

$X = \sqrt{\alpha P}X_S + \sqrt{(1-\alpha)P}X_B$

At the receiver, successive interference cancellation (SIC) or joint decoding recovers the streams, followed by task-aware semantic decoding ($g_S$) and bit-decoding ($g_B$). The system can flexibly allocate transmit power between semantic and bit channels by tuning $\alpha \in [0,1]$ (Ahmed et al., 13 Jun 2025).

This hybrid design supports dynamic adaptation and multiplexing of users with distinct traffic profiles (e.g., sensor updates—semantics, files—bits).

3. Multiple Access and Resource Allocation in SemCom Systems

Efficient multiple access in mixed SemCom/BitCom regimes leverages power-domain non-orthogonal multiple access (NOMA). For a 2-user uplink with ordered channel strengths $|h_1|^2 > |h_2|^2$:

• Received superposition:

$$Y = \sqrt{\alpha_1 P} h_1 X_1 + \sqrt{\alpha_2 P} h_2 X_2 + N$$

• SINR for user $i$:

$$\gamma_1 = \frac{\alpha_1 P |h_1|^2}{\alpha_2 P |h_2|^2 + N_0}$$

$\gamma_2 = \frac{\alpha_2 P |h_2|^2}{N_0}$

• BitCom rate: $R_{B,i} = W\log_2(1+\gamma_i)$
• SemCom rate: $R_{s,i} = \frac{W I}{K L}\epsilon(K,\gamma_i)$

Optimal resource allocation problems arise, e.g., maximizing expected semantic rate $E[R_{s,2}]$ subject to $E[R_{B,1}]\ge \bar R$ via power fraction and on-off policies (Ahmed et al., 13 Jun 2025). Such optimization intrinsically couples semantics-driven objectives with classic capacity constraints.

In multi-user contexts, reinforcement learning (RL) driven resource allocation and semantic compression model (SCM) selection further optimize trade-offs among semantic accuracy, latency, and energy, introducing metrics such as rate-distortion efficiency (RDE) and leveraging proximal policy optimization (PPO) for policy learning under non-convex constraints (Lin et al., 23 Jun 2025).

4. Multi-Modal and Context-Aware Semantic Communication

Advanced SemCom frameworks extend beyond unimodal semantic extraction to multi-modal data fusion and context-sensitive adaptation:

• Multi-modal Fusion: Individual modality-specific encoders (text, image, audio, etc) produce semantic embeddings $S_{\mathrm{text}}$, $S_{\mathrm{img}}$, $S_{\mathrm{audio}}$, that are fused $$S_{\mathrm{joint}} = \Phi_\mathrm{fuse}([S_{\mathrm{text}} ; S_{\mathrm{img}} ; S_{\mathrm{audio}}])$$, transmitted, and decoded with task-aware heads (Ahmed et al., 13 Jun 2025).
• Unified Multi-task (U-DeepSC) Architectures: These dynamically select sub-networks per task/modality, effectively implementing joint source-channel coding (JSCC) per type, combined in a shared latent space (Ahmed et al., 13 Jun 2025).

Context-aware architectures exploit LLM-powered gating and mixture-of-experts (MoE) encoders to select which features to transmit, as in the CaSemCom framework. Here, context embeddings from channel/task states control both content masking and expert activation, maximizing semantic yield per bandwidth while reducing retransmission overhead (Liu et al., 29 May 2025).

5. Semantic Communication in Satellite and Non-Terrestrial Networks

SemCom is especially effective in satellite and non-terrestrial communication scenarios characterized by severe bandwidth limitations and challenging channel conditions. In satellite links:

• Channel Model: Free-space pathloss $$L_{\rm FS}(d) = 20 \log_{10} (\frac{4\pi d f_c}{c})$$, equivalent gain $\rho(d) = \rho_0 (d_0/d)^\beta$.
• Semantic Bandwidth Gain: Since practical semantic similarity $\epsilon(K,\gamma)<1$, the requisite channel bandwidth to achieve a target semantic rate $R_s$ can be reduced by $1/\epsilon$ relative to bit-based approaches, delivering substantial spectrum savings (Ahmed et al., 13 Jun 2025).
• Semantic Relays and Knowledge Base Coordination: Semantic relays act as amplify-and-forward nodes at the semantic level, periodically re-aligning user equipment (UE) knowledge bases via dedicated KB coordinator satellites to ensure semantic interoperability (Ahmed et al., 13 Jun 2025).

System-level performance can be further optimized by intelligent KB caching and power allocation, maximizing semantic secrecy throughput in presence of potential eavesdroppers sharing semantic KBs (Liu et al., 21 Apr 2025).

6. Open Research Directions and Systemic Challenges

Several key open problems and research frontiers are identified:

• AI-Native Physical Layers: End-to-end co-design of semantic encoders with new physical-layer waveforms via deep learning.
• Knowledge Base Management: Versioning, secure distribution, and standardization of shared KBs is nontrivial, given the need for semantic alignment across diverse devices and vendors.
• Energy Efficiency: Investigation of trade-offs between encoder/decoder complexity and power consumption, especially for constrained IoT and satellite devices.
• Semantic Security & Interpretability: Defending against adversarial semantic tampering or leakage, as well as ensuring explainability and privacy of semantically encoded content.
• Cross-layer Protocols: Design of medium access and routing protocols natively aware of semantic metadata and quality-of-semantic (QoS) constraints, e.g., semantic-aware scheduling for minimizing semantic age-of-information or optimizing value-of-information.

The integration of these principles into the design of scalable, robust, and efficient semantic-aware networks remains pivotal for the realization of 6G and beyond (Ahmed et al., 13 Jun 2025).